Skip to main content
gpt-image-2-all / 2.5-all / 2.6-all do not need their own skill. They share the single gpt-image-2 series skill with gpt-image-2 (official) and gpt-image-2-vip — they all use the same OpenAI Images API; only --model differs. For the full install, SKILL.md, and script, see GPT-Image-2 Series Agent Skill.

2.6-all also shares this skill (and is the default model)

gpt-image-2.6-all is the 2.6-generation reverse line. It uses the same main skill, and it is now the skill’s default model — omit --model and you get it:
The biggest difference from 2.5-all: 2.6-all really honors size (native 512–1536; 2K/4K are super-resolution tiers — common sizes exact, mismatch returns 502), and it supports n 1–4 plus server-side direct URLs; quality still only accepts auto/medium/high with no verified difference, and mask is not guaranteed to be local. See GPT-Image-2.6-All Overview for the full boundaries.

What this model is good for

Fastest

Reverse ChatGPT line, ~30–60s — the fastest of the three channels.

Cheapest

Flat $0.025/image regardless of size/quality — great for volume.

T2I / fusion

Text-to-image and up to 16-image fusion (no mask inpainting).

How to use it in the skill

After installing the gpt-image-2 series skill, set --model to gpt-image-2.5-all (or gpt-image-2-all; same price and behavior):
To make it the default channel (no --model each time), add a line to gpt-image-2/.env:
These red lines apply to 2.5-all / 2-all only (2.6-all honors size; do not mix the 2.5 conclusions into it).Red lines: gpt-image-2.5-all / gpt-image-2-all do not accept size / quality / n —
  • put size/ratio in the prompt (e.g. “portrait 2:3”, “16:9 banner”); a passed size is ignored or errors;
  • passing n>1 still returns only 1 image but bills per count.
The skill script already omits size/quality/n for this model, so normal use of --model gpt-image-2-all is safe; only watch out if you hand-craft requests outside the script. For locked size/4K use gpt-image-2-vip; for quality tiers / mask use official gpt-image-2.